version: "1.0.0" name: guide-prompting-os-06-evaluation-and-optimization description: Use the Evaluation and Optimization guide from docs/prompting-os/06-evaluation-and-optimization.md as a bounded, public-safe workflow. category: guides source:

• docs/prompting-os/06-evaluation-and-optimization.md

Evaluation and Optimization

Trigger

Use this skill when the task is about docs/prompting-os/06-evaluation-and-optimization.md or the topic covered by Evaluation and Optimization, and the source guide is the right local authority to inspect before acting. Do not use this skill for unrelated work or as permission to broaden the task beyond the files and actions the human requested.

Purpose

Convert the source guide into a short, repeatable operating procedure: read the authoritative document, keep work scoped, avoid unsupported claims, and report evidence clearly.

Inputs

• The user's current task or question.
• The source file: docs/prompting-os/06-evaluation-and-optimization.md.
• Any files, tools, or outputs explicitly named by the user.
• The verification commands that apply to the touched area.

Scope

Allowed: read the source file, follow its public-safe workflow guidance, and edit only files that the current task explicitly brings into scope.

Forbidden unless explicitly approved: secrets, .env files, credentials, private links, private local paths, dependency additions, workflow YAML edits, destructive commands, direct pushes to protected branches, and broad rewrites outside the current task.

Procedure

1. Read docs/prompting-os/06-evaluation-and-optimization.md in full before editing, summarizing, or applying it.
2. Read AGENTS.md if the task may change repository files.
3. Restate the specific goal, in-scope files, out-of-scope files, and verification commands.
4. Apply only the guidance relevant to the current task; link back to the source guide for deeper explanation instead of copying large sections.
5. Mark fast-changing product behavior, model availability, pricing, platform support, or setup details as official-doc verification items unless freshly checked.
6. Run the verification listed below if any repository files changed.

Verification

For repository edits, run the strongest relevant subset of:

python scripts/repo_health_check.py
python scripts/safe_autofix.py --check
python -m unittest discover -s tests

For read-only use, verify by citing the source file and clearly separating source facts from assumptions or recommendations.

Failure Cases

• The source file is missing, unreadable, or no longer matches the task.
• The task needs current external product behavior that has not been verified

in official docs.

• The requested change would touch secrets, private data, dependencies,

workflow YAML, or destructive commands without explicit approval.

• Verification fails for a reason outside the current task's scope.
• The source guide and the human's request conflict; ask for direction instead

of guessing.

Final Report

• Source used: docs/prompting-os/06-evaluation-and-optimization.md.
• Files changed, if any.
• Commands run and whether they passed.
• Current-claim or official-doc verification items left unresolved.
• Remaining risks or skipped checks.

Disable Path

Remove skills/guide-prompting-os-06-evaluation-and-optimization/ from the installed harness location, or remove this folder from the repository together with its skills/manifest.json entry and skills/INDEX.md row in the same reviewed change.